[H]ardOCP HSF Review Changes & Feedback.

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,532
After our latest HSF article here, we wanted to tackle some of the issues you guys brought up.


OK, looking at the changes here.

  • Use Stock Out of Box Fans & [H] Supplied Fan for tests
  • Add sound level testing
  • Higher wattage CPU - Will use a Yorksfield Core 2 Quad QX960 / OCing of course added.
  • No more software temp solutions - Hardware Sperry DT-506 thermometer
  • Grind channel in Heatspreader for thermistor placement
  • Use new standard Asus P5K Deluxe Motherboard
  • Add cost of fan to heatsink cost (if fan not included)


Please give us your feedback, both barrels.
 
As Kyle has mentioned above in the *change log* using a quad will be great. If a hs/f can endure good temps on a quad, then it becomes a no brainer that it will fare well with a dual core.
 
* Use Stock Out of Box Fans & [H] Supplied Fan for tests I disagree, a control fan should be used for an equal CFM going through the fins. However, should a high speed one be used? Moderate? Low?

* Add sound level testing
* Higher wattage CPU - Will use a Yorksfield Core 2 Quad QX960 / OCing of course added.

* No more software temp solutions - Hardware Sperry DT-506 thermometer If it's accurate across tests. I don't know how it would be configured but would mounting/remounting affect its performance? I've always though Intel's TAT is a benchmark core temp reader

* Grind channel in Heatspreader for thermistor placement See above
* Use new standard Asus P5K Deluxe Motherboard
* Add cost of fan to heatsink cost (if fan not included)
 
* Use Stock Out of Box Fans & [H] Supplied Fan for tests I disagree, a control fan should be used for an equal CFM going through the fins. However, should a high speed one be used? Moderate? Low?

* Add sound level testing
* Higher wattage CPU - Will use a Yorksfield Core 2 Quad QX960 / OCing of course added.

* No more software temp solutions - Hardware Sperry DT-506 thermometer If it's accurate across tests. I don't know how it would be configured but would mounting/remounting affect its performance? I've always though Intel's TAT is a benchmark core temp reader

* Grind channel in Heatspreader for thermistor placement See above
* Use new standard Asus P5K Deluxe Motherboard
* Add cost of fan to heatsink cost (if fan not included)

As mentioned a control fan setup will be used as well as out of the box fan if supplied. To address your controller suggestion, RPM is hardly the only variable to CFM though, blade pitch and surface area impact that figure as well.

Mounting should not impact this test any more than any other. I am thinking of filling the channel with epoxy and lapping after the thermister is placed. This should allow for an "identical" mating surface on every mount. Taking the software out of the equation removes another variable that we have to deal with.
 
Ah I see, I wasn't aware you'd be using a control fan and the stock fan. I understand that RPM isn't everything, as well. What I meant by deciding what control fan to use is that some heatsinks are better optimized to use a fan that is rated at a relatively high CFM. I remember reading that a Scythe Ninja outperforms other heatsinks with more densely packed fins using a low CFM fan, making it more ideal for quiet systems. I think I read that on SPCR some time ago.
 
Won't software give the same #s as long as its the same software being used and the same hardware (besides the HS)? Not sure if someone gave a detailed explanation in that other thread or not as to why using the software temp #s is a bad idea.
 
Ah I see, I wasn't aware you'd be using a control fan and the stock fan. I understand that RPM isn't everything, as well. What I meant by deciding what control fan to use is that some heatsinks are better optimized to use a fan that is rated at a relatively high CFM. I remember reading that a Scythe Ninja outperforms other heatsinks with more densely packed fins using a low CFM fan, making it more ideal for quiet systems. I think I read that on SPCR some time ago.


We do not have the resources to be looking into specifically which fans on the market (out of hundreds) are better for specific heatsinks.
 
Won't software give the same #s as long as its the same software being used and the same hardware (besides the HS)? Not sure if someone gave a detailed explanation in that other thread or not as to why using the software temp #s is a bad idea.


The fact is that some people do not trust software. Removing the software from the equation gives us less variables. Talking to Intel now to see if we can get the latest version of their Thermal Analysis Tool in order to give us another certified way of monitoring CPU heat.

I do know I can trust a high dollar thermometer placed in between the mating surfaces that is statically mounted into the face of the heatspreader. I will post some pics when I get done with the mod. :)
 
# Add sound level testing
# Higher wattage CPU - Will use a Yorksfield Core 2 Quad QX960 / OCing of course added.

Yes and Yes!

I don't think hardware monitoring is required, so long as temperatures in IntelTAT and coretemp coincide... I think the process of grinding and placing the thermistor on all heat sinks would be MORE variable than a software solution...
 
I think the process of grinding and placing the thermistor on all heat sinks would be MORE variable than a software solution...

Not if I fix the thermister in place with high temp epoxy then lap the final product flat. It will never move and the mating surface of the CPU will never change.

But you know what, if enough of you tell me that Intel TAT and coretemp is enough, we will go with that too.
 
I suggest choose a CPU that pulls 200W or more under load when max oc'ed on air, e.g. a 65nm quad, a 5xx series prescott or a smithfield. I don't think a 45nm quad will be that hungry even at 4GHz.
 
Well you could always test one heatsink with the hardware tester and run TAT/coretemp as well, then modify the fan speed and see how both the software and hardware read the temp change. If they match up or are at least very close (both in original, final, and change) then you know the software is good and you won't need to use the hardware :)


Well then the hardware modifications will have already been made to the $1000 CPU....
 
I suggest choose a CPU that pulls 200W or more under load when max oc'ed on air, e.g. a 65nm quad, a 5xx series prescott or a smithfield. I don't think a 45nm quad will be that hungry even at 4GHz.


We kicked that around, but considering that yorkfield is the future and will be scaling down quickly, that is what we are going to focus on.
 
Oh I meant doing it prior to modifying the cpu if you could. It probably wouldn't be as accurate as there would be a slight gap but probably close enough to compare the temps. Or insert it in the heatsink if thats possible (one not all)?


Thanks for the suggestion.
 
Why the Zalman paste and not AS5 ? I use IC D7 personally,and think its better then both but still.And TAT ? Everyone I know of here and elsewhere use Coretemp 94 or 95
but whatever is done hardware temp monitoring is the only way to go I would think on a Pro hardware site such as this...Also,as has been already brought up,the chip used was way to low on the TDP scale to even be considered IMHO.A mainstream quad or QX model shuld be the standard.

Only a quad,and only in a closed case.Also room temps must be very very striclty controlled,as ambient can radically effect the outcome with this kind of thermal testing.
 
Why the Zalman paste and not AS5 ? I use IC D7 personally,and think its better then both but still.And TAT ? Everyone I know of here and elsewhere use Coretemp 94 or 95
but whatever is done hardware temp monitoring is the only way to go I would think on a Pro hardware site such as this...Also,as has been already brought up,the chip used was way to low on the TDP scale to even be considered IMHO.A mainstream quad or QX model shuld be the standard.

Only a quad,and only in a closed case.Also room temps must be very very striclty controlled,as ambient can radically effect the outcome with this kind of thermal testing.

They explained the zalman vs AS5.

First, zalman has the nice brush
Second, Zalman doesn't need time to get to its best temps like the AS5, which adds days to a review when waiting for it to cure.
 
Why the Zalman paste and not AS5 ? I use IC D7 personally,and think its better then both but still.And TAT ? Everyone I know of here and elsewhere use Coretemp 94 or 95
but whatever is done hardware temp monitoring is the only way to go I would think on a Pro hardware site such as this...Also,as has been already brought up,the chip used was way to low on the TDP scale to even be considered IMHO.A mainstream quad or QX model shuld be the standard.

Only a quad,and only in a closed case.Also room temps must be very very striclty controlled,as ambient can radically effect the outcome with this kind of thermal testing.


Thanks for the feedback, both your questions are answered in the article already on TIM and room temp.
 
I want to see a roundup of the smaller coolers, as I mentioned, and gave a list of, in theother thread.
 
I want to see a roundup of the smaller coolers, as I mentioned, and gave a list of, in theother thread.

When I picked these I picked them because they represented the "bigass" coolers that were just coming out in the summer. We will certainly look at a wider variety of products. Promise. :)
 
Not to sure if this has been said or not. Could you list the temperatures of the mosfets and possibly ram (maybe even northbridge since lots of boards connect the mosfets and northbridge) when you have a downward blowing cooler versus tower cooler? I would like to know how big of an impact removing the downward blowing air has on these parts. I know my purchase of a cooler be affected if I find out that my tower is going to cause my mosfets to overheat.

Edit: Also could you see that if changing the orientation of the heat pipes has any affect on the temperatures? I know people say it doesn't but I would like to see some testing on whether or not the orientation (you have the boards flat on a bench while they will be sideways in a pc) has an affect.
 
What about "included" thermal paste. That would be a complete, out-of-box expirence. I know, everyone and their grandmother uses AS5 or something better, but again, that's added cost usually to the HSF.
Just a thought.
 
A quick read of the Intel CPU data sheet will confirm this method (link below) of reading the on-board CPU DTS is both correct and 100% accurate: And shows that Coretemp if put in the "Delta to Tjunction" mode is correct as well. Since you will be using a "gold standard" cpu questions about incorrect math due to assumed Tjunction(max) in Coretemp are not applicable and the number of deg C to assertion of the PROCHOT# interupt contained in the cpu MSR would be an very accurate way of comparing heatsink performance.

http://www.hardforum.com/showpost.php?p=1031080147&postcount=5

---------------------------------------------

An appendix to the Thermal and Mechanical Guidelines from Intel will outline the exact procedure including all needed dimensions for installing the thermal probe. I like your lapping idea as opposed to the Intel method of shaving the excess epoxy off. As this is the documented Intel method of determining Tcase for the thermal load lines I assume you are using it as a guide ?

I would like to see both SW and hardware) , and if you can persuade Intel to get a new TAT out, woot /cheer that would be great. Good show.

----------------------------------------

hmm Check for heasink mating area bring flat ?

Nothing fancy just one of these laid on the heatsink mating surface with a strong light behind would be interesting and gently slide it along the surface looking for areas where the light was most visible and snap a pic. guick and dirty but very informative IMO. This is just a variation of the razor blade test with a known flat tool. $77 for a pair. Come in handy for checking your lapp job too.

Page 2214 on http://www.mcmaster.com/

Steel Parallels
Excellent for inspection and layout on surface plates and for setups on drill presses, milling and grinding machines, and shapers. These parallels are hardened and ground tool steel. Made in the USA.

Parallels Ground on Four Sides

Use these parallels when you need more than two parallel sides. Pairs are parallel, straight, and matched on width and height within 0.0001" for 6" and 8" lengths, 0.0002" for 12" lengths, and 0.0003" for longer lengths

6 incher should do it. a taller thin one would help occlude the backlight and make taking the picture and showing imperfections easier.


Wd. and Ht.
Lg. Wd. Ht. Tolerance Square Per Pair

6" 1/4" 1/2" ±0.0002" 0.0005" 2259A12 $77.00

6" 1/4" 3/4" ±0.0002" 0.0005" 2259A14 77.00

6" 3/8" 3/4" ±0.0002" 0.0005" 2259A17 77.00

6" 1/2" 3/4" ±0.0002" 0.0005" 2259A19 84.00

A standard disclaimer that this is a test of one individual heatsink and manufacturing tolerances make it important to check your own heatsink and minor imperfections should not be considered a indication of overall heatsink quality. However large deviations from flat would tend to indicate imprecise manufacturing methods or finish and require the end user to be especially vigilant in checking their heatsink and correcting the imperfections if it was deemed important.
 
Personally I'd use both hardware and software to measure the temps. Coretemp is supposed to be 100% accurate due to reading the digital info from the processor (it doesn't read an analog temp diode which is why it doesn't work with older processors) and as BP pointed out there are settings to ensure it's accuracy. Thing is though it doesn't reflect the temp for the entire core, just the area that the temps are being polled which is why the hardware would come in handy, it would give a more comprehensive look at the temp of the surface of the chip. Bear in mind it won't be totally accurate as you're going to be dealing with losses from multiple interfaces. First the thermal interface between the cores and the IHS and then the thermal interface of the diode and the IHS, thermally conductive epoxy and the epoxy with the IHS.

The diode will be getting X heat from the IHS where it directly contacts it while the epoxy will be contributing Y heat since it's a different thermal material than the IHS and will heat at a different rate as well as have different thermal conductivity than the IHS. This is on top of the fact that the IHS will be cooled faster than the CPU cores due to it being in direct contact with the HSF. The upside is that the epoxy might actually act as an insulator between the base of the HSF and the diode helping to insure that the core temps are more accurately represented by it as opposed to the temperature of the base of the HSF.
 
Back
Top